Michael ParkerFebruary 24, 2013

What is CouchDB?● One of those hipster NoSQL databases● Document-oriented, not relational

○ No joins, prefer denormalization● RESTful API for all operations● Views add structure via secondary indexes● Other perks:

○ Multiversion concurrency control (lock-free)○ MapReduce framework○ Great web admin interface○ Multi-master replication

Document-oriented● Like a key-value store, flat namespace

○ In CouchDB, key is called a document identifier, or doc id

● Documents stored in JSON format○ "Schemaless," but the schema resides in your app○ JSON also format for HTTP body

● Similar: MongoDB (no relation), Redis, Cassandra

JSON Document in CouchDB{ "_id": "emp_001", "_rev": "1-4c6114c65e295552ab1019e2b046b10e", "name": { "first": "Dante", "last": "Hicks", }, "phone": "310-555-1212", "interests": ["philosophy", "Star Wars"]}

RESTful URLs● All resources in database identified by URL● Basic URL structure:

○ /db_id: A database○ /db_id/doc_id: A document in a database

● URL components and JSON fields starting with an underscore are special, e.g.:○ /_config: CouchDB configuration parameters○ /db_id/_all_docs: A cursor across all

documents○ /db_id/_design/design_doc: A design

document for a database

RESTful Methods● HTTP methods for CRUD actions

○ GET: Read data, typically a document○ HEAD: Like GET without a body, typically used to

check if a document exists○ PUT: Creates new databases, documents, and other

resources○ POST: Updates these resources○ DELETE: Deletes these resources

RESTful Status Codes● HTTP status codes for server responses

○ 200 OK: Request completed successfully (e.g. retrieving, updating, deleting documents)

○ 201 Created: Resource created (used with PUT)○ 202 Accepted: Request completed and operation

pending (e.g. for background operations)○ 401 Unauthorized: Bad username or password○ 404 Not Found: Resource missing○ 409 Conflict: MVCC failure, or concurrent

modification to a document○ 500 Internal Server Error: Everybody panic

Creating a Database● Request (abbr.):

PUT /new_db/ HTTP/1.1

● Response (abbr.):HTTP/1.1 201 Created

{"ok": true}

Creating a Database● Request (abbr.):

GET /_all_dbs HTTP/1.1

● Response (abbr.):HTTP/1.1 200 OK

["new_db"]

Creating a Document● Request (abbr.):

PUT /my_db/emp_001 HTTP/1.1

{ "name": { "first": "Dante", "last": "Hicks", }, "phone": "310-555-1212", "interests": ["philosophy", "Star Wars"]}

Creating a Document● Response (abbr.):

HTTP/1.1 201 Created

{ "ok": true, "id": "emp_001", "rev": "1-4c6114c65e295552ab1019e2b046b10e"}

Retrieving a Document● Request (abbr.):

GET /my_db/emp_001 HTTP/1.1

Retrieving a Document● Response (abbr.):

HTTP/1.1 200 OK

{ "_id": "emp_001", "_rev": "1-4c6114c65e295552ab1019e2b046b10e", "name": { "first": "Dante", "last": "Hicks", }, "phone": "310-555-1212", "interests": ["philosophy", "Star Wars"]}

Retrieving a Document● Request (abbr.):

GET /my_db/missing_emp_007 HTTP/1.1

● Response (abbr.):HTTP/1.1 404 Object Not Found

{ "error": "not_found", "reason": "missing"}

Views● Secondary indexes for querying by other

than _id● Written in JavaScript, executed with Mozilla

SpiderMonkey engine● Defined in the design document

Views● Define view emps_by_interest:

function(emp_doc) { for (var i = 0; i < emp_doc.interests.length; ++i) { var interest = emp_doc.interests[i]; emit(interest.toLowerCase(), null); }}

● Request (abbr):GET /my_db/_design/my_dd/_view/emps_by_interest?key=philosophy HTTP/1.1

Views● Response (abbr):

HTTP/1.1 200 OK

{ "total_rows": 1, "offset": 0, "rows": [ {"id": "emp_001", "key": "philosophy", "value": null} ]}

● Append &include_docs=true in request to return documents with results

Multiversion Concurrency Control (MVCC)● Every document has a _rev attribute

○ Only required field other than _id● Ensures that client is updating latest data

○ No accidental clobbering● Lock-free concurrency control● If multiple clients attempt to write

concurrently, exactly one succeeds every time○ Always making "forward progress"

Multiversion Concurrency Control (MVCC)

GET id=doc_id

_rev=X, v=1

GET id=doc_id

_rev=X, v=1

t

_rev=X, v=2_rev=Y, v=2

_rev=X, v=2

HTTP 409

GET id=doc_id

_rev=Y, v=2

_rev=Y, v=3

HTTP 200, _rev=Y

_rev=Z, v=3HTTP 200, _rev=Z

CLIENT 1 CLIENT 2

Benchmarking

https://github.com/mgp/iron-cushion

● Setup:○ Server: Intel Core 2 2.83GHz quad-core, 4GB RAM○ Client: 1.83 GHz Intel Core Duo MacBook○ 100Mbit LAN, 100 concurrent connections○ first, bulk insert 2,000,000 documents○ second, intersperse 20,000 create and read

operations, 30,000 update and delete operations● Caveat: no indexes

BenchmarkingbulkInsertRate: 10,003.030 docs/sec

createProcessingRate: 949.141 docs/secreadProcessingRate: 9,015.862 docs/secupdateProcessingRate: 980.172 docs/secdeleteProcessingRate: 980.154 docs/sec

Thanks!http://couchdb.apache.org/

michael.g.parker@gmail.comhttps://github.com/mgp

http://mgp.github.com/couchdb-la-hn.pdf